| --- |
| library_name: pytorch |
| license: other |
| tags: |
| - real_time |
| - bu_auto |
| - android |
| pipeline_tag: other |
|
|
| --- |
| |
|  |
|
|
| # RangeNet-Plus-Plus: Optimized for Qualcomm Devices |
|
|
| RangeNet-Plus-Plus (also stylized as RangeNet++) projects a LiDAR point cloud onto a 5-channel range image (depth, x, y, z, intensity) and applies a DarkNet-53 encoder with a decoder head to predict per-point semantic class labels in real time. |
|
|
| This is based on the implementation of RangeNet-Plus-Plus found [here](https://github.com/PRBonn/lidar-bonnetal). |
| This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/rangenet_plus_plus) library to export with custom configurations. More details on model performance across various devices, can be found [here](#performance-summary). |
|
|
| Qualcomm AI Hub Models uses [Qualcomm AI Hub Workbench](https://workbench.aihub.qualcomm.com) to compile, profile, and evaluate this model. [Sign up](https://myaccount.qualcomm.com/signup) to run these models on a hosted Qualcomm® device. |
|
|
| ## Getting Started |
| There are two ways to deploy this model on your device: |
|
|
| ### Option 1: Download Pre-Exported Models |
|
|
| Below are pre-exported model assets ready for deployment. |
|
|
| | Runtime | Precision | Chipset | SDK Versions | Download | |
| |---|---|---|---|---| |
| | ONNX | float | Universal | QAIRT 2.42, ONNX Runtime 1.24.3 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/rangenet_plus_plus/releases/v0.53.1/rangenet_plus_plus-onnx-float.zip) |
| | TFLITE | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/rangenet_plus_plus/releases/v0.53.1/rangenet_plus_plus-tflite-float.zip) |
|
|
| For more device-specific assets and performance metrics, visit **[RangeNet-Plus-Plus on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/rangenet_plus_plus)**. |
|
|
|
|
| ### Option 2: Export with Custom Configurations |
|
|
| Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/rangenet_plus_plus) Python library to compile and export the model with your own: |
| - Custom weights (e.g., fine-tuned checkpoints) |
| - Custom input shapes |
| - Target device and runtime configurations |
|
|
| This option is ideal if you need to customize the model beyond the default configuration provided here. |
|
|
| See our repository for [RangeNet-Plus-Plus on GitHub](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/rangenet_plus_plus) for usage instructions. |
|
|
| ## Model Details |
|
|
| **Model Type:** Model_use_case.driver_assistance |
| |
| **Model Stats:** |
| - Model checkpoint: darknet53_rangenet++ |
| - Input resolution: 64x2048 |
| - Input channels: 5 |
| - Number of output classes: 20 |
| - Backbone: DarkNet-53 |
|
|
| ## Performance Summary |
| | Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |
| |---|---|---|---|---|---|--- |
| | RangeNet-Plus-Plus | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 41.39 ms | 3 - 335 MB | NPU |
| | RangeNet-Plus-Plus | ONNX | float | Snapdragon® 8 Elite Mobile | 58.534 ms | 0 - 329 MB | NPU |
| | RangeNet-Plus-Plus | ONNX | float | Snapdragon® X2 Elite | 49.501 ms | 101 - 101 MB | NPU |
| | RangeNet-Plus-Plus | ONNX | float | Snapdragon® X Elite | 100.677 ms | 100 - 100 MB | NPU |
| | RangeNet-Plus-Plus | ONNX | float | Snapdragon® X Elite | 100.677 ms | 100 - 100 MB | NPU |
| | RangeNet-Plus-Plus | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 74.569 ms | 0 - 457 MB | NPU |
| | RangeNet-Plus-Plus | ONNX | float | Qualcomm® QCS8550 (Proxy) | 102.668 ms | 3 - 5 MB | NPU |
| | RangeNet-Plus-Plus | ONNX | float | Qualcomm® QCS9075 | 159.07 ms | 2 - 8 MB | NPU |
| | RangeNet-Plus-Plus | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 58.534 ms | 0 - 329 MB | NPU |
| | RangeNet-Plus-Plus | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 44.421 ms | 0 - 315 MB | NPU |
| | RangeNet-Plus-Plus | TFLITE | float | Snapdragon® 8 Elite Mobile | 60.235 ms | 0 - 296 MB | NPU |
| | RangeNet-Plus-Plus | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 78.37 ms | 0 - 511 MB | NPU |
| | RangeNet-Plus-Plus | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 595.836 ms | 0 - 308 MB | NPU |
| | RangeNet-Plus-Plus | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 97.484 ms | 0 - 96 MB | NPU |
| | RangeNet-Plus-Plus | TFLITE | float | Qualcomm® SA8775P | 154.76 ms | 0 - 308 MB | NPU |
| | RangeNet-Plus-Plus | TFLITE | float | Qualcomm® SA8775P | 154.76 ms | 0 - 308 MB | NPU |
| | RangeNet-Plus-Plus | TFLITE | float | Qualcomm® SA8775P | 154.76 ms | 0 - 308 MB | NPU |
| | RangeNet-Plus-Plus | TFLITE | float | Qualcomm® QCS9075 | 167.36 ms | 0 - 107 MB | NPU |
| | RangeNet-Plus-Plus | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 195.519 ms | 1 - 500 MB | NPU |
| | RangeNet-Plus-Plus | TFLITE | float | Qualcomm® SA7255P | 595.836 ms | 0 - 308 MB | NPU |
| | RangeNet-Plus-Plus | TFLITE | float | Qualcomm® SA8295P | 171.967 ms | 0 - 302 MB | NPU |
| | RangeNet-Plus-Plus | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 60.235 ms | 0 - 296 MB | NPU |
|
|
| ## License |
| * The license for the original implementation of RangeNet-Plus-Plus can be found |
| [here](https://github.com/PRBonn/lidar-bonnetal/blob/master/LICENSE). |
|
|
| ## References |
| * [RangeNet++: Fast and Accurate LiDAR Semantic Segmentation](https://ieeexplore.ieee.org/document/8967762) |
| * [Source Model Implementation](https://github.com/PRBonn/lidar-bonnetal) |
|
|
| ## Community |
| * Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI. |
| * For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com). |
|
|
